Monocular Visual Odometry and Obstacle Detection System Based on Ground Constraints

摘要

The realization of visual odometry and obstacle detection system with only a single camera is proposed. Though the rotation and translation parameters of frame-to-ftame motion can be extracted from essential matrix using SVD decomposition, the absolute scale of the translation cannot be derived from monocular motion estimation. The scale ambiguity problem is solved by applying constraints based on the planar assumption and the known mounting of the camera. A region-based obstacle detection method is proposed in this paper. Firstly, the image is segmented into regions. Then whether a region is on the ground is determined according to three criteria: homography constraint, feature points distribution and boundary points' reconstruction. Practical experimental results show that the proposed method successfully estimate the robot location and find the ground plane with only a single camera.